Recurrent Neural Networks for Modelling Gross Primary Production

TL;DR

This paper evaluates recurrent neural network architectures for estimating daily Gross Primary Production, demonstrating comparable performance among models and highlighting the importance of radiation and remote sensing data, especially during climate extremes.

Contribution

It provides a comparative analysis of RNN, GRU, and LSTM architectures for GPP estimation, emphasizing the significance of remote sensing data and model performance during climate extremes.

Findings

LSTMs outperform in predicting climate-induced GPP extremes

All models show similar performance for full-year and growing season predictions

Incorporating radiation and remote sensing inputs improves GPP prediction accuracy

Abstract

Accurate quantification of Gross Primary Production (GPP) is crucial for understanding terrestrial carbon dynamics. It represents the largest atmosphere-to-land CO$_2$ flux, especially significant for forests. Eddy Covariance (EC) measurements are widely used for ecosystem-scale GPP quantification but are globally sparse. In areas lacking local EC measurements, remote sensing (RS) data are typically utilised to estimate GPP after statistically relating them to in-situ data. Deep learning offers novel perspectives, and the potential of recurrent neural network architectures for estimating daily GPP remains underexplored. This study presents a comparative analysis of three architectures: Recurrent Neural Networks (RNNs), Gated Recurrent Units (GRUs), and Long-Short Term Memory (LSTMs). Our findings reveal comparable performance across all models for full-year and growing season predictions. Notably, LSTMs outperform in predicting climate-induced GPP extremes. Furthermore, our analysis highlights the importance of incorporating radiation and RS inputs (optical, temperature, and radar) for accurate GPP predictions, particularly during climate extremes.